[26 19:48:23] <haskellbridge> <M​an of Letters> second backport down (https://github.com/haskell/cabal/pull/10858); very fast, because CI skipped a lot of checks, probably erroneously, probably geekosaur noticed that before
[26 19:50:56] <haskellbridge> <g​eekosaur> Doesn't seem like mergify's going to fix it, we'll have to drop the docs hack to get full ci
[26 19:56:09] <haskellbridge> <g​eekosaur> or find an alternative to mergify that still maintains all our label trickery
[26 19:58:13] <haskellbridge> <g​eekosaur> for those playing along from home: mergify pushes a PR in as soon as all conditions are satisfied, not waiting for CI to finish. the docs hack skips the build checks and marks them successful, by ignoring changes to non-docs files (which is the only option provided by github actions triggers)
[26 19:59:25] <haskellbridge> <g​eekosaur> github's own merge train support doesn't allow us to use cooldowns, which is somewhat problematic for how we do things
[26 20:16:02] * merijn ([email protected]) has joined
[26 20:18:39] <haskellbridge> <M​an of Letters> geekosaur: I'm not clear on how the docs hack affects the situation; e.g., the PR in question didn't have the marker that it's doc changes only
[26 20:19:47] <haskellbridge> <g​eekosaur> that is in fact the problem. there are two versions of the validate run: one which ignores changes to non-doc files and returns success, and the normal one which builds and validates eveyrhting and returns success
[26 20:20:03] <haskellbridge> <g​eekosaur> the docs one, since it doesn't build or validate, always returns success first
[26 20:20:32] <haskellbridge> <g​eekosaur> mergify takes that as meaning that, since validate "succeeded", it's okay to merge
[26 20:21:25] <haskellbridge> <g​eekosaur> you will see that GHA itself doesn't, since changes to PRs don't count as validation-succeeded until the actual validate run finishes successfully
[26 20:22:12] <haskellbridge> <g​eekosaur> (which is what was tested before the docs hack went in)
[26 20:23:16] <haskellbridge> <g​eekosaur> an additional complication is that mergify uses the configuration on master, so you can only test mergify configuration by merging changes and seeing what happens on the next PR that it is told to merge
[26 20:23:40] <haskellbridge> <M​an of Letters> nasty; since no solution is on the horizon, I agree, let's rip out the docs-only validator
[26 20:24:29] <haskellbridge> <g​eekosaur> there might be alternatives, such as using the publicly-available marge-bot (same as ghc ci uses) if it supports the cooldown stuff